Deep latent variable models

Lieu: CNAM
Date et Heure de début: 20-04-2018

Le prochain séminaire de Statistique appliquée du CNAM se tiendra le  vendredi 20 avril de 11h à 12h en salle 31.2.85 (2 rue Conté, entrée 31, 2eme étage). Le séminaire est ouvert et vous êtes tou.te.s les bienvenu.e.s.

Nous accueillerons Pierre-Alexandre Mattei (IT Université de Copenhage) pour une conférence intitulée :

Deep latent variable models

Abstract :Deep latent variable models combine the approximation abilities of deep neural networks and the statistical foundations of generative models. The induced data distribution is an infinite mixture model whose density is extremely delicate to compute. Variational methods are consequently used for inference, following the seminal work of Rezende, Mohamed, and Wierstra (ICML 2014) and Kingma & Welling (ICLR 2014). We will provide a general review of these models and techniques, viewed from a statistical perspective. In particular, we will study the well-posedness of the exact problem (maximum likelihood) these variational approaches approximatively solve. We show that most unconstrained models used for continuous data have an unbounded likelihood. This ill-posedness and the problems it causes are illustrated on real data. We also show how to insure the existence of maximum likelihood estimates, and draw useful connections with nonparametric mixture models. Furthermore, we describe an algorithm that allows to perform missing data imputation using the exact conditional likelihood of a deep latent variable model. On several real data sets, our algorithm consistently and significantly outperforms the usual imputation scheme used within deep latent variable models.

Organise: MSDMA
Contact: Avner Bar-Hen